(The following is a slightly edited conversation that took place on Facebook recently, on Roberta D’Alessandro’s page. There’s also one comment by Roberta. It’s reproduced here with permission.)
Martin Haspelmath (Reacting to a Facebook comment that it’s hard to understand the syntax of human languages): Syntax suddenly starts working if it’s framework-free! But I admit it may not be so cool…
Richard Larson: Same is true for physics! It suddenly “starts working” if you toss out all this silly theorizing about forces, particles and least effort principles! 🙂
Martin Haspelmath: Have you read my paper about framework-free theory, Richard? I’d be curious to hear what you’d say about it. I find the reasoning quite compelling, and I’d like to hear counterarguments. Some additional discussion (including of the notorious physics analogy) is found in an earlier post on this blog.
Richard Larson: There is no analogy to physics being drawn here, notorious or otherwise. (I could have chosen chemistry, geology bio, etc.). The point is that, at least in the way I’m understanding “framework”, “framework-free syntax (semantics, phonology, etc.)” has the same status as “theory-free science”. It’s a non-sequitur. I suspect you have a narrow characterization of “framework”, and I suspect it is connected with what you do as a typologist – the equivalent in our field of an observational astronomer. But we’ll see.
Martin Haspelmath: What I’m saying that the best way to do linguistic science is not by constructing a uniform framework for all languages, but by constructing a different framework for each language, and by looking for adaptive explanations of the similarities across languages. This is very far from being “theory-free” – but I think that we need different kinds of theories for particular languages and for language in general.
Roberta D’Alessandro: You see, no framework free grammar, but a multitude of frameworks!
Martin Haspelmath: Yes! I chose the provocative title “framework-free” partly to have a conversation-starter, but also because most people seem to simply assume that there must be a single framework for all languages. This would be perfectly possible (and maybe partially it is the case, in some way), but it’s not the only possibility. Humans are able to create highly specific cultural frameworks, so why shouldn’t they be able to create original linguistic frameworks that differ from those of other languages?
Richard Larson: Martin, I think what you are endorsing is something very classically Chomskian, viz. the autonomy of syntax. On the LSLT view, grammar is a purely formal instrument that humans make use of to express thought. It doesn’t come with an interpretation and it doesn’t dictate an interpretation. On this view although the human grammatical mechanism will provide expressions, structures, features, etc., there is in fact no guarantee (and no real expectation) that the interpretation of those constructs will be the same across languages: no universal set of phonetic features, no universal set of theta-features (or roles), no universal hierarchies of adverbs, no universal set of grammatical categories, etc. What will be universal will be structural, not contentful. The opposite view to LSLT is generative semantics, either of the classical decompositional sort, whose modern descendent is DM, or the neo-Montagovian sort which sees structure as arising out of semantic type (which is supposed to follow out of meaning, somehow.). Chomsky held that there is a single structural capacity or schematism underlying all languages. But that does not entail any more parallelism than what that shared capacity entails. It does not entail that humans will make the same use of that capacity in building their languages. Give people all across the world brick and mortar making capacity. How similar do you expect their buildings to be?
Martin Haspelmath: Yes, exactly. The similarities between the diverse frameworks (a different one for each language) need not be due to the stuff from which they are made: functional selection can do the trick. Give people wood and sawing capacity, and their buildings will still look very similar. Thus no rich UG is needed (I don’t know if this is classically Chomskian, but it’s certainly minimalistic).
Richard Larson: I think one shouldn’t overdo the 1945 structuralism in this. If the suggestion is that languages can differ endlessly on all dimensions then we most definitely part company. I still haven’t seen a functional explanation yet for why linear order inverting rules of the sort Chomsky talked about at the beginning aren’t deployed, why displacement affects constituents, etc. There are definite constraints imposed in the picture Chomsky set out. It’s the interpretation of the formatives that is not anchored. But I agree that there is greater potential for structural variation than is usually countenanced.
Martin Haspelmath: I agree that some things are probably hard-wired, but I don’t see how the usual practice of generative grammar could help us find out which of them are. It seems that you’re not terribly happy about that usual practice either, so there may indeed be some unexpected convergence here… It seems that your position is similar to “substance-free phonology”, as advocated, e.g., by Roberta’s coauthor Tobias Scheer. I’d say that UG is substance-free, but given that so much of language structure is substance-based, how do we find out what is in UG?
Richard Larson: You misunderstand me. I don’t have any complaints about current practice, the core of which is the detailed formal analysis of specific languages. It is from those investigations that the underlying algebra, the formal operations and constraints, will become apparent. (And by the way, I don’t think terminology like “substance-free” is an improvement over Chomsky’s original characterization of syntax as “autonomous”). How do we find out what is UG? By being perceptive and creative. It takes these to look at the rotations and reflections on a square and the Warlpiri subsection system and see behind both the multiplication table for the dihedral group of order 8 – the autonomous abstract structure of which those two are just interpretations. Is there is a mechanical procedure that can be embedded in our methodology that would have derived this? Of course not. We’re doing science. My two cents.
Martin Haspelmath: I agree that perceptiveness and creativeness might possibly lead to a discovery of pre-established patterns (=the framework common to all languages). But how do we know whether we have made such a discovery, as opposed to just one more proposal? I think it will have to come from converging evidence on the same proposal, but I don’t see converging evidence – just more and more loosely connected proposals. Where we have convergence, it seems to be on cross-linguistic generalizations, as recently highlighted by Peter Svenonius and by Roberta D’Alessandro in Barcelona. The question how to explain these seems to be as wide open as always. (And the generalizations themselves are best found in the framework-free mode.)